Introduction

Global warming has resulted in a large-scale retreat of glaciers throughout the world (Oerlemans 2005). The evidence for this is particularly strong in high-altitude areas, such as the central Himalaya (Solomon et al 2007; Hambrey et al 2009; Quincey et al 2009; Ye et al 2009), where widespread recession is evident from the rapid growth in the number and size of glacial lakes (Chikita et al 2001; Quincey et al 2007; Bolch et al 2008; Tartari et al 2008; Wang et al 2008; Ye et al 2009).

Deglaciation processes are also revealed by the amount of suspended solids transported by glacier waters melting into the lake (Østrem et al 2005), with consequences for light propagation in the water body. If glacier water influx into a lake increases, then the maximum suspended particle size and particle number density will increase, and this affects light scattering. For decreased meltwater input, absorption because of water increases and the preferential red absorption because of water are enhanced. In lakes that have no glacial input, lake water is clear, and there is almost no scattering or absorption because of suspended matter (Kargel et al 2005). The recognition of lakes hydraulically connected to the glacier hydrological systems, hence full of silt (eg moraine-dammed supraglacial lakes), as opposed to those that are isolated (eg cirque lakes), would add value to studies (eg Tartari et al 2008) on processes of glacial-lake formation and expansion. Ultimately, these studies might contribute to an assessment of glacial lake hazard, because some moraine-dammed lakes may be unstable and potentially susceptible to sudden discharge of large volumes of water and debris, which could cause floods hazardous to communities and infrastructure downstream (Richardson and Reynolds 2000; Kattelmann 2003; Bajracharya et al 2007; Quincey et al 2007; Bolch et al 2008).

Besides indicating the dynamics of the interaction between glaciers and lakes, suspended solids play a fundamental role in the aquatic system itself. They regulate the transport routes of all types of materials and contaminants in aquatic ecosystems (Wetzel 1983) and ultimately determine the primary productivity of water (Zhang et al 2008). As an indicator of water clarity (eg Secchi disk depth and water transparency, see Håkanson et al 2007), suspended solid concentration is also a macrodescriptor of water quality directly related to many variables of general use in lake management (Baban 1999). Water quality monitoring in the Himalayan region would contribute to proper management of some glacial lakes that could provide valuable water, energy, and tourism resources to local residents in the downstream regions (Komori 2008; Salerno et al 2008).

Suspended solids are traditionally measured by collecting water samples and then analyzing them in the laboratory. However, the extreme climate variations and the inaccessible terrain of the Himalayan region make it difficult to perform an assessment of suspended solids with traditional in situ methods. Remote sensing represents a useful tool for augmenting or replacing in situ methods and can be used to survey large areas regularly for dynamic monitoring.

Suspended solids are one of the parameters that can be measured successfully by means of remote sensing in inland waters (Lindell et al 1999 and references therein). Generally, 2 approaches can be used (Dekker et al 1995; Cracknell et al 2001) to estimate water quality parameters (including suspended solids) in lakes from satellite data:

Quantitatively, the relationships developed to assess water quality within semiempirical approaches are sensor-dependent and may only apply to the data from which they are derived. Instead, well-calibrated and validated bio-optical models within the analytical approach are more general and may be applicable to every scene acquired over selected lakes independently of ground measurements. However, a great effort is needed to establish robust relationships between IOPs and concentrations of water quality parameters, because lakes are optically complex waters in which the relations between IOPs and concentrations of water components seem to have local-regional behavior (Kutser et al 2001) and may vary over time (Giardino et al 2007).

The purpose of this study was to measure the capability of satellite data to retrieve the suspended particulate matter (SPM) concentrations of glacial lakes in the Mount Everest region. The semiempirical approach was considered suitable to demonstrate the value of this technique in rapidly mapping SPM in lakes and studying the dynamics of interaction between glaciers and lakes. Moreover, the relations between the absorption coefficients of particles and CDOM with water quality parameters (ie SPM and CDOM), all derived from in situ measurements, were described in the perspective of bio-optical modeling. This is a preliminary step for implementing a remote-sensing–based procedure for water quality monitoring of Himalayan lakes.

Study area, data sources, and methods

The present investigation concerns the glacial lakes in the Himalayan region that extend to the north and south of Mount Everest at approximately 27°58′N; 86°45′E. The study area covers approximately 1300 km² and presents unique features, being surrounded on all sides by the highest mountain range on Earth. Within this area, 17 lakes were surveyed in September–October 2008 (Figure 1, Table 1). The lakes located in Nepal are indicated with the same numbers used by the lake cadastre of Sagarmatha National Park (SNP) (Tartari et al 1997; Tartari et al 2008), whereas, for the Chinese lakes, an arbitrary numeration was used. The lakes were distinguished in 3 categories, supraglacial, moraine-dammed, and cirque, according to both Tartari et al (2008) and field notes.

Figure 1

The study area in Himalaya. The map was developed based on an ALOS AVNIR-2 image acquired on 24 October 2008. The location of the sampled lakes are indicated by numbers: for the lakes sampled to the south of Mount Everest the cadastre numbers in Tartari et al (1997 and 2008) were used. The map also indicates main glaciers, peaks, and valleys in the Nepal–China border region.

Table 1

Characteristics of the sampled lakes with indication of their origin. (Table extended on next page.)

In the southern part (Nepal), the surveyed area is demarcated by the valley of the rivers Bhote Koshi, Dudh Koshi, and Imja Khola (Figure 1), which together drain into the Dudh Kosi river and afterward into the Ganges on the plain near Chatra. The surface area of the region is 1025 km² and covers 89% of the 1148 km² of SNP, located in the east of Nepal in a complex transition zone between the high Himalaya and Tibet. The region is characterized by different geological units (Bortolami 1998), where the surface occupied by the lakes is very small compared with that occupied by glaciers, but the conditions are very different from valley to valley. In particular, 5 lakes (nos. 121, 128, 129, 136, and 138; Figure 1) were located in the Bhote Koshi Valley, which drains many small sub-basins in the western part of SNP and where more than 10 glacier formations are present. Of these 5 lakes, 2 are moraine-dammed lakes located in the immediate vicinity of the glacier (nos 136 and 138, Figure 1), 2 others are cirque lakes (nos 128 and 129, Figure 1), without any extensive glacier in their watersheds; the last lake (no. 121, Figure 1) is a small lake fed by a glacier located about 200 m above it. The other 3 lakes (nos. 71, 75, and 76; Figure 1) were located in the Dudh Koshi Valley, which is drained by the 25-km-long Ngozumpa Glacier. This valley contains an interesting group of later moraine-dammed lateral lakes. The remaining 5 lakes (nos. 10, 24, 31, 32, and 161; Figure 1) were located in the valley of Imja Khola where geomorphological conditions are more diversified. Of these, 3 were cirque lakes (nos. 10, 31, and 32; Figure 1). The fourth lake was Imja Lake (no. 161, Figure 1), a moraine-dammed lake, which began to grow after 1962 (Bolch et al 2008), fed by the Imja and Lhotse Shar glaciers. Toward the west, the large moraine-dammed Chola Lake (no. 24, Figure 1) was sampled. The lake drains the south side of Lobuche Peak and the northern icy side of the Cholatse and Taboche mountains.

In the northern part (Tibet, China), the study area is located in Dingri County, where the unique and various geographic conditions and the sensitive and vulnerable environment make it a perfect site for investigating changes in water quality and the dynamic process of glacial lakes in the context of climate change. Most of the lakes are located in the valleys close to the glaciers. They are formed by the accumulation of vast amounts of water from the melting of snow and ice cover and by blockage of end moraines. In particular, the investigated lakes were located in the Rongbuk river catchment, on the northern slopes of Mount Everest. The Rongbuk River is fed by the Rongbuk Glacier, which is the largest glacier in the catchment (Ye et al 2009). Many supraglacial lakes developed on the glacier terminus. The sampled lakes include 4 supraglacial lakes (nos. EL 2–1, EL 2–2, EL 2–3, and EL 2–4; Figure 1), among them the largest (EL 2–1, 1223 km², Figure 1 and Table 1) is also moraine dammed.

Satellite data

Satellite data were obtained by the Advanced Visible and Near Infrared Radiometer type 2 (AVNIR-2) onboard ALOS, a Japanese Earth Observation satellite launched in January 2006. AVNIR-2 is a multispectral radiometer for observing land and coastal zones with a 10-m spatial resolution, a 70-km swath width (at nadir), and a revisiting time of 46 days. It has 3 bands in the visible (VIS) range (wavelength centers of 420, 520 and 610 nm; bandwidths of 80 nm) and 1 band in the near-infrared (NIR) (wavelength center 760 nm, bandwidth 130 nm). AVNIR-2 data investigated in this study were acquired on 24 October 2008 (Scene ID ALAV2A146473040), just at the end of fieldwork activities, in a cloudless sky. The image was acquired at the end of the monsoon season before the first big snowfall. Image data were orthorectified and corrected for the atmospheric effects by using the 6S code (Vermote et al 1997). The code was run according to nadir viewing and illumination conditions of the sun's zenith of 43° and the sun's azimuth of 156°. The aerosol optical depth at 550 nm was set to the mean value of the month (0.02) derived from the Ev-K2-CNR AERONET station. A threshold of band 4 to band 1 ratio was applied for binary discrimination to classify water pixels from the other surfaces. The threshold was chosen according to the spectral properties of water in terms of brightness (lower reflectance in comparison with other surfaces) and shape (lower reflectance in the NIR than in the blue wavelengths). The cast shadow areas wrongly classified as water body (commission error of 7.2%) were removed by visual inspection.

The 6S-derived reflectances are showed in Figure 2 for the lakes sampled to the south of Mount Everest. Three groups of spectra with increasing degrees of brightness are clearly distinguished with consequences of water color. By integrating the water reflectance in the whole range of AVNIR-2 bands, from VIS to NIR wavelengths (R_VIS-NIR), it can be observed that higher reflectance spectra (R_VIS-NIR > 20%) correspond to lakes whose color, in the pseudo true color AVNIR-2 image, appears gray (nos. 71, 121, 136, and 161). However, lower reflectance spectra (R_VIS-NIR < 10%) correspond to lakes whose color appears dark blue (nos. 10, 31, 32, 128, and 129). In between (10% ≤ R_VIS-NIR ≤ 20%), there is a group of spectra that correspond to lakes whose color appears turquoise (nos. 24, 75, 76, and 138). This can also be confirmed visually (Figure 2, embedded photos).

Figure 2

Satellite-derived reflectance for the lakes sampled to the south of Mount Everest, in Nepal. 3 groups of spectra are distinguishable for increasing brightness; lower-reflectance spectra are indicated with blue lines; higher-reflectance spectra are indicated with light gray lines; in between are the medium-reflectance spectra. The embedded photos show the variation of colors for 3 lakes from dark blue to gray (in situ: top 3 photos; crop from AVNIR-2 image: 2 images below).

Results and discussion

Suspend particulate matter

Widely variable SPM concentrations were encountered in the study area (Table 1); the maximum (320 g/m³) and minimum (0.41 g/m³) SPM concentrations were measured in lakes EL 2–1 and no. 129, to the north and to the south of Mount Everest, respectively. Overall, in cirque lakes the SPM concentrations were lower than 3 g/m³, whereas supraglacial and moraine-dammed lakes were characterized by wide variations in SPM concentrations, with values greater than 100 g/m³ in the 2 largest sampled lakes (nos. 161 and EL 2–1). However, a few exceptions were observed: in cirque lake no. 121, the SPM concentration was about 20 g/m³, whereas in the supraglacial lake EL 2–3 in China the SPM concentration was only 0.5 g/m³. Field notes indicated that lake no. 121 was receiving water from a glacier located approximately 200 m above it; whereas the supraglacial lake EL 2–3 was a small, shallow, and clear pond above the Rongbuk Glacier.

The SPM concentrations measured in lakes in Nepal (see Figure 2 and Table 1) were used to establish the relationship with satellite data. The AVNIR-2 derived reflectance R_VIS-NIR was considered the dependent variable. This rough estimation of lake albedo was considered suitable to represent the intensity of water reflectance and the observed desaturation of lake colors from dark-blue to gray. The relationship between SPM and R_VIS-NIR is shown in Figure 3A; symbols are used to distinguish the samples for increasing R_VIS-NIR values according to 3 clusters (Figure 2): R_VIS-NIR < 10%, 10% ≤ R_VIS-NIR ≤ 20%, and R_VIS-NIR > 20%. A direct correlation between SPM and R_VIS-NIR with a high degree of fit (R²  =  0.921) was retrieved, thus the regression line (Figure 3A) was applied to each lake pixel of AVNIR-2 R_VIS-NIR data to generate a map of SPM concentrations for all the whole lakes in the study area. The satellite-derived SPM concentrations were compared with in situ data, measured in the 4 lakes located in the northern part of the study area (nos. EL 2–1, EL 2–2, EL 2–3, and EL 2–4; Figure 1). A high degree of fitting was found (R²  =  0.924) (Figure 3B), despite an underestimation of lower values and an overestimation of higher concentrations detected. The results confirm the capacity of the technique to describe a wide range of SPM concentrations and to quickly locate the lakes according to the 3 classes of SPM concentrations (dark blue, turquoise, and gray).

Figure 3

(A) Natural logarithmic of SPM concentrations versus AVNIR-2 atmospherically corrected reflectance data (R_VIS-NIR). The regression line was used to map SPM concentrations from satellite data in the whole image; (B) comparison of satellite-derived and in situ measured SPM concentrations for the 4 lakes in China. In both plots symbols change according to water colors (hence to water reflectance and SPM concentrations): diamonds for dark blue lakes; squares for turquoise lakes; triangles for light gray lakes.

The map of SPM for a spatial subset of the study area is shown in Figure 4A (see box in Figure 1), in the region of the Imja Kola valley. The SPM concentrations were mapped with color codes comparable with SPM concentration ranges, which also correspond to the brightness of water reflectance: in blue, the lakes in which SPM is lower than 3 g/m³ (and R_VIS-NIR < 10%); in turquoise, the lakes in which SPM ranges from 3–6 g/m³ (10% ≤ R_VIS-NIR ≤ 20%); in gray, the lakes in which SPM concentration is greater than 16 g/m³ (R_VIS-NIR > 20%). The map shows a sequence of several lakes whose SPM concentrations varied significantly. In the western part, the 2 cirque lakes nos. 31 and 32 (see Figure 1 and Table 1) were characterized by low concentrations of SPM (≤3 g/m³). When moving toward the east, 2 lakes with significantly dissimilar SPM concentrations are distinguishable. In Amphulaptse Lake (circular shape) the concentrations of SPM were lower than 3 g/m³, which indicated the absence of glacier–lake interactions. This behavior was also described by Bolch et al (2008), who did not observe any change in the lake's size from 1962–2003. Finally, the large Imja Lake (no. 126, Figure 1) was characterized by higher concentrations of SPM, which indicated that it has been receiving silt inputs from the glacier. The result was in agreement with the progression of Imja Lake, whose growth has been carefully analyzed (Yamada 1998; Quincey et al 2005; Bajracharya et al 2007; Bolch et al 2008; Hambrey et al 2009). For Imja Lake, the SPM map was further investigated by defining more-appropriate color ranges for appreciating the spatial distribution of suspended solids (Figure 4B). The SPM concentrations mainly ranged between 75 and 125 g/m³, with higher concentrations observed toward the northern part, whereas any significant pattern was instead visible, moving from the glacier tongue toward the dam. The satellite-derived SPM concentrations were comparable with Chikita (2004): in his study, in situ measured suspended sediment concentrations at the surface were about 80 g/m³ (they reached 100 g/m³ toward the lake bottom), which suggests that the silt inputs from the glacier did not vary significantly.

Figure 4

(A) Map of SPM concentrations in lakes located in Imja Kola valley (see box in Figure 1). In dark blue the lakes with SPM < 3 g/m³; in turquoise the lakes with SPM from 3–16 g/m³, in light gray the lakes with SPM > 16 g/m³. The largest lake on the map is Imja Lake (no. 161, Figure 1), the 2 small lakes in the eastern part are 2 cirque lakes (nos. 31 and 32, Figure 1) sampled during the fieldwork activities. The circular lake south of Imja Lake with low SPM concentration is Amphulaptse Lake. (B) Zoom on Imja Lake where SPM concentrations are mapped with a color table describing the spatial variation of SPM within the lake.

Absorption coefficients

The values of the normalized a_p(440)/a_tot(440) and a_CDOM(440)/a_tot(440) [with a_tot(440)  =  a_p(440) + a_CDOM(440)] for the samples collected in 13 lakes in Nepal are shown in Table 2. The statistics are computed according to the 3 groups of lakes, previously defined according to R_VIS-NIR values and SPM ranges: dark blue, turquoise, and gray lakes. The average value of a_p(440) is greatest in turquoise lakes (0.29), whereas the average value of a_p(440) reaches the minimum (0.052) in dark-blue lakes. On the contrary, the average normalized value of a_CDOM(440) was higher in the dark-blue lakes (0.948) than in turquoise lakes (0.710). Although the range of variability did not vary significantly, these results suggest that, in the clearest lakes, the main absorption component was CDOM, whereas, in turquoise lakes, it was particles. The gray lakes stay in between, which indicated that CDOM and particles equally contributed to the absorption of light.

Table 2

Global and subcluster statistics for the absorption coefficients at 440 nm of particles a_p(440) and CDOM a_CDOM(440) with respect to the total absorption coefficients a_tot(440).^a)

Similarly, all sampled lakes present a small range of variability with regard to the slopes of both particle (S_p) and CDOM (S_CDOM); these slopes being the scalar used to model the well-known (eg Babin et al 2003) exponential function: a_i(λ)  =  a_i(440)e^−Si, where i represents the particle “p” and CDOM, respectively. As showed in Table 3, the average values of S_p and S_CDOM, are 0.0116 and 0.0093, respectively; the latter being comparable with S_CDOM values found both in lakes (Strömbeck and Pierson 2001; Ma et al 2006) and in coastal waters (Babin et al 2003).

Table 3

Global and subcluster statistics for the spectral slopes of particle and CDOM.^a)

Similarly, all sampled lakes present a small range of variability with regard to the slopes of both particle (S_p) and CDOM (S_CDOM); these slopes being the scalar used to model the well-known (eg Babin et al 2003) exponential function: a_i(λ)  =  a_i(440)e^−Si, where i represents the particle “p” and CDOM, respectively. As showed in Table 3, the average values of S_p and S_CDOM, are 0.0116 and 0.0093, respectively; the latter being comparable with S_CDOM values found both in lakes (Strömbeck and Pierson 2001; Ma et al 2006) and in coastal waters (Babin et al 2003).

The absorption coefficients of a_p(440) and a_CDOM(440) were then analyzed to find their relationship with SPM and S_CDOM, respectively, under the perspective of parameterizing a bio-optical model for the study area. The relationship between a_p(440) and SPM concentrations (see Table 1), whose strength (R²  =  0.965) indicates that it can be applied in a bio-optical model to relate the absorption coefficient of particles to SPM concentrations, is shown in Figure 5A. With respect to CDOM, we found no significant results in investigating the inverse relationship that may exist between a_CDOM(440) (see Table 1) and S_CDOM (Carder et al 1989; see Figure 5B). Therefore, in implementing a bio-optical model, the relationship between and a_CDOM(λ) and a_CDOM(440) should be a function of a constant slope S_CDOM (Dekker et al 2001). The results suggest that both SPM and CDOM could be obtained by inverting the bio-optical model, even though in situ data would be necessary to investigate the back-scattering coefficient of particles, which is needed to fully parameterize the bio-optical model. The technique then would simultaneously provide an assessment of SPM and CDOM (Brando and Dekker 2003).

Figure 5

(A) Scatterplot of a_p(440) as a function of SPM for the 13 lakes, sampled in Nepal; (B) scatterplot of S_CDOM as a function of a_CDOM(440) for the 13 lakes, sampled in Nepal.

Conclusions

This study demonstrated how ALOS AVNIR-2 satellite data may be used to retrieve SPM concentrations in high altitude lakes: the lakes are characterized by wide variations in SPM concentrations, ie from less than 1 g/m³ to more than 300 g/m³. The semiempirical approach was considered suitable to assess SPM concentrations in the study area. A robust relationship was developed (R²  =  0.921) by comparing atmospherically corrected reflectance data, integrated in the VIS and NIR wavelength range, with SPM concentrations, sampled in 13 lakes south of Mount Everest (Nepal). The satellite-derived SPM concentrations were comparable with in situ data (R²  =  0.924) collected in 4 lakes north of Mount Everest (Tibet, China). The results confirmed the capacity of this technique to retrieve a wide range of SPM concentrations and to quickly locate the lakes according to 3 classes of SPM concentrations. In 2 large ice-contact lakes (no. 161, Imja in Nepal, and EL 2–1 in China), the amount of silt was found to be particularly high (SPM > 100 g/m³), which indicated that the lakes are hydraulically connected to the glaciers. The results are in agreement with recent findings describing the dramatic retreat of the Rongbuk and Imja glaciers, which has resulted in the rapid expansions of the EL 2–1 (Ye et al 2009) and Imja (Bolch et al 2008) lakes, respectively.

The study was complemented by the analysis of absorption coefficients of particles and CDOM. The results indicated a direct correlation between SPM and absorption of particles, with an almost complete absence of particles in clear-blue waters where the main absorption component was instead CDOM. The absorption coefficients were investigated as a preliminary step toward the goal of bio-optical modeling, which would allow remote sensing–based techniques to assess SPM concentrations independently of ground measurements (Zhang et al 2008). Moreover, the technique would simultaneously provide an assessment of CDOM, whose impact on water quality is not secondary, being a major structuring component of lake ecosystems and protecting the aquatic biota from ultraviolet solar radiation (Kutser et al 2005). The availability of a bio-optical model, together with time-series satellite images, normalized by illumination and atmospheric effects (as AVNIR-2 image used in this study), would make it possible to assess water quality in such a remote area synoptically and at different periods. Nevertheless, operational sensors with the spatial resolution of ALOS AVNIR-2 but improved temporal resolution are needed for implementing continuous monitoring programs. Such a requirement could be fulfilled by the recently launched RapidEye Satellite with a spectral configuration comparable with AVNIR-2.

The technique presented in this study might help to explain the processes of glacial lake formation and expansion, with emphasis on those processes characterized by rapid growth in the number and dimensions of glacial lakes that may generate catastrophic outburst floods (Kattelmann 2003). Being a macrodescriptor of water quality, the assessment of SPM in glacial lakes of the Himalayan region might also be of interest for resource use in the downstream region. Water quality measurements (both SPM and CDOM) in Himalayan lakes might also receive increasing attention in the wider scientific community. High-altitude lakes are regarded as potentially sensitive ecosystem indicators of global change because of their cold and dilute abiotic environment, low biodiversity, and considerable water clarity ranges, poor functional redundancy, and relative lack of local human perturbation (Skjelkvåle and Wright 1998; Sommaruga 2001; Battarbee et al 2002; Psenner et al 2002; Rogora et al 2008).

Open access article: please credit the authors and the full source.